Live circularity dashboard
Session totals and every scan in real time, each with an item ID, recovery value, CO₂ avoided and C2C tier.
CIRCULA
Saudi ArabiaCIRCULA is a Saudi circular-economy startup that helps industrial operators turn waste into recoverable value, and we built its platform end to end. A camera on a Qualcomm RB3 Gen 2 edge board classifies each scrap item at the sorting point (PCB boards, solar PV fragments, battery cells) on the device itself, so it keeps working where connectivity is poor. The cloud platform expands each item into its material composition and scores it against the Cradle to Cradle v4.1 framework: recovery rate, recovery value in SAR and CO₂ avoided. A second path takes an operator’s own ERP, SCADA and lab extracts, maps their columns once, validates and de-duplicates every row, and builds a material-flow baseline per stream in which every figure links back to its source record. Roles, tenant isolation, an audit trail and full English and Arabic make it ready for industrial clients.
- Manufacturing
- Edge AI
- Computer Vision
- Circular Economy
- Data Engineering
- Web App
- AI Solutions
02 · PROJECT OVERVIEW
Project overview
A hybrid edge-to-cloud platform for CIRCULA: on-device classification of scrap on a Qualcomm RB3 Gen 2, deterministic Cradle to Cradle scoring in the cloud, and a data layer that turns ERP, SCADA and lab extracts into an auditable material-flow baseline. Built spec-first from a proof of concept into a multi-tenant, bilingual product.
- BUDGET
- $60K
- TIMEFRAME
- 5.5 months, ongoing
- RELEASE DATE
- 2026
- TEAM SIZE
- 3 members
- PLATFORM
- Web · Edge device
- TECHNOLOGIES
- Next.js
- FastAPI
- PostgreSQL
- Redis
- Qualcomm RB3 Gen 2 (QNN)
- YOLO
- MQTT
- Python
03 · PROBLEMS & SOLUTIONS
Problems & Solutions
CIRCULA needed one platform that works at a sorting point with no reliable link, explains every number it shows, and meets an industrial client’s rules for data, roles and audit, in English and Arabic.
Edge AI // On-device classificationSorting decisions were made by eye on remote sites
Solution: A YOLO detector runs on the Qualcomm RB3 Gen 2 NPU and classifies each item at the sorting point; detections wait on the device and sync when the link returns.
Deterministic scoring // Cradle to Cradle v4.1A scanned item said nothing about its metals
Solution: Each item type expands through a Bill of Materials into its materials, scored on five C2C categories where the weakest sets the tier, with Recovery %, SAR value and CO₂.
Data ingestion // ERP · SCADA · LabPlant data sat in ERP, SCADA and lab files
Solution: Upload an Excel or CSV extract, map its columns to the CIRCULA data model once, save the mapping as a template, and validate and import every row in the background.
Material flow // Baseline per streamNo one could say where a stream’s material went
Solution: A baseline per stream and month (inputs, stock, waste, resources consumed, recoverable) where every figure carries a confidence rating and opens its source records.
Governance // Roles · Audit · ArabicIndustrial data needs roles, audit and Arabic
Solution: Four roles with per-tenant isolation, an audit trail of every change, append-only operator corrections, and every screen in English and Arabic, laid out right to left.
04 · FROM CONCEPT TO COMPLETION
From concept to completion
How we transformed ideas into impactful results.
PHASE 1 OF 6
Discovery phase
Scoped the proof of concept with CIRCULA: three item classes, a hybrid edge-to-cloud design and a live-demo date. Later rounds turned CIRCULA’s Top-5 priorities into dated delivery gates.
FEATURES WE IMPLEMENTED
Features we implemented
Material analysis and next steps
Score any product on demand and get recovery opportunities and next steps derived from its materials.
Material records with their source
Every imported row as a record with its stage, confidence, file and row number, filterable and correctable.
Lifecycle stages and flow figure
A stage strip and a flow figure drawn to scale show where material sits; every band opens its records.
Roles, users and audit trail
Admins issue accounts, set roles and read who changed what, and when, across the organisation.
05 · IMPACT CREATED
Impact created
16 weeksFirst pilot
From the first commit on 13 May 2026 to a working pilot shown at LEAP 2026 on 31 August.
+40%Defect detection
More defective and mis-sorted items caught at the sorting point once the edge model classifies every item.
AutomatedForecasting
Recovery and material-flow forecasts built from the operator’s own ERP, SCADA and lab data instead of spreadsheets.
Rare metalsTracking
Silver, copper, silicon and other valuable metals in PCB boards, solar fragments and battery cells tracked from the sorting point.
4,200+Automated tests
About 3,000 backend and 1,250 frontend tests, plus 20 end-to-end suites, run on every change.
100%Traceable figures
Every baseline figure opens the source records it was summed from, down to the file and row.
Want to achieve similar results? Let’s develop your idea!
07 · CASE STUDIES
See more of our projects
Case study · Maritime cybersecurityReal-time protection for IT and OT systems at sea
Fleets ran on more and more digital systems at sea – communication, IT and OT – with no centralized cyber protection, low bandwidth on board and legacy systems that were hard to secure.
- WHAT IT COST
- $80K
- TIME TO LIVE
- 8 months
Case study · CIS payrollSales & lead re-engagement
Their CRM was a mess – no analytics, no calendar sync, conversations not linked to leads. Working the existing base – chasing, qualifying, booking calls – was done entirely by hand.
- WHAT IT COST
- $4–6K
- TIME TO LIVE
- 3.5 weeks
Case study · Healthcare payrollMulti-system payroll automation
Two accountants spent ~2 weeks each cycle: entering medics' hours by hand, chasing doctors for sign-off, then paying everyone manually.
- WHAT IT COST
- $11K
- TIME TO LIVE
- 10 weeks